CTCN Talk: Wanyu Lei

Feature Computation and Representation in the Brain and Artificial Neural Networks

Wanyu Lei, PhD Student, Yale University

Understanding how neural networks compute and represent features is a crucial pursuit in both neuroscience and artificial intelligence. In this talk, I will delve into my research efforts on feature processing in both biological and brain-inspired artificial neural networks. Regarding biological networks, I will present our findings concerning visual feature computation in the early visual system — the retina. I will highlight our discovery of a novel group of orientation-selective wide-field amacrine cells in the mouse retina and discuss how different network motifs and intrinsic biophysical properties permit orientation selectivity at an algorithmic level. In the realm of artificial neural networks, I will present our work investigating the connection among perturbation robustness, capacity, and manifold geometry of brain-inspired power-law neural networks. Taken together, these studies provide theoretical insights into how the brain is optimized to perform feature-processing tasks.